solver.com

If you want to solve optimization problems larger than these limits, or if you want even faster solution times on problems within these limits, you can expand the Solver Platform SDK with field-installable Solver engines.

Each Solver engine supports the same easy-to-use API (application programming interface) as the Solver Platform SDK.  With one line of code, you can add the plug-in Solver to the collection of available engines -- then set its parameters and use it to solve a problem.  You define your model, solve problems, write Evaluators (callback functions), and access the solution in the same way for all Solver engines -- so it's easy to test several Solver engines to find out which one performs best on your problem.

Each Solver engine also "plugs into" the Premium Solver Platform -- our flagship product for optimization in Microsoft Excel -- and is licensed for dual use with both the Solver Platform SDK and the Premium Solver Platform.

The following Solver Engines are currently available.  Check back frequently, since Frontline Systems is adding even more Solver engines to extend the flexibility and power of the Solver Platform SDK product line!

Linear and Quadratic Programming Problems

You can also use large-scale nonlinear Solver engines with the Premium Solver Platform to solve large-scale LP problems, but the above Solver engines offer much better performance on such problems.  All of the above Solver Engines are typically very fast on large-scale LP problems.  Note: Solution of large-scale QP problems using any of these Solver Engines may be limited by available memory.

Conic Optimization Problems

The MOSEK Solver is designed for conic optimization, and offers the best performance on SOCP problems.  The Large-Scale GRG Solver, Large-Scale SQP Solver and KNITRO Solver are designed to solve both convex and non-convex NLP problems, but they also handle second order cone (SOC) constraints.

Integer and Constraint Programming Problems

Every Solver engine for the Premium Solver Platform will handle problems with integer variables, including variables subject to the "alldifferent" constraint.  If you have a large or challenging mixed-integer or constraint programming problem, however, the Large-Scale LP Solver may be faster, and the XPRESS Solver may be fastest on these problems.

Smooth Nonlinear Optimization Problems

The MOSEK Solver can handle very large smooth convex NLP problems, but it does not support non-convex problems.  The Large-Scale SQP Solver can solve very large smooth convex and non-convex NLP problems, but its practical upper limit on the degrees of freedom (i.e. the number of variables minus the number of constraints that are binding at the solution) is about 2,000.  Thanks to both interior point methods and active-set methods, the KNITRO Solver can handle the largest smooth convex and non-convex NLP models, and the number of degrees of freedom can be much larger than 2,000.

You can also use Solver engines designed for global and nonsmooth optimization with the Premium Solver Platform to solve smooth NLP problems, but the above Solver engines offers better performance on such problems.

Global Optimization Problems

The OptQuest Solver finds global solutions and also handles nonsmooth problems, but it has no test for global optimality.  The other Solver Engines use the Premium Solver Platform's multistart or clustering methods to seek all locally optimal solutions, and select the best of these as the probable globally optimal solution.  Although these Solver Engines accept large or unlimited size problems, the practical limit for global optimization problems is much lower -- comparable to the OptQuest Solver.

Nonsmooth Optimization Problems

The OptQuest Solver is designed for nonsmooth optimization, and usually offers the best performance on arbitrary Excel models, especially if they include integer variables.  The Large-Scale SQP Solver integrates the Evolutionary Solver and is very effective for problems with some nonsmooth, and other smooth and linear, variable occurrences -- but the practical limit on nonsmooth variables and constraints is much lower than for smooth problems.

You can try Solver engines designed for smooth non-convex nonlinear optimization on nonsmooth problems, but they may not successfully deal with nonsmooth or discontinuous functions that are important to the model.

Solver Platform SDK Product Overview